Platform Models for Sustainable Internet Regulation · of each ISP, their IP platform is a single-firm platform, while the Internet is a multi-firm platform that happens to use

Platform Models for Sustainable Internet Regulation

kc claffy and David D. [email protected] and [email protected]

August 15, 2013

Abstract

The dynamic nature of the telecommunications industry, with its rapidly changing tech-nology and industry structure, presents a serious challenge to the theory and practice of reg-ulation, which has a slower time scale and a tendency to embed assumptions about technol-ogy and industry into regulation. This paper proposes a model that attempts to capture twodurable and persistent features of today’s telecommunications ecosystem: the use of layeredplatforms to implement desired functionality; and interconnection between actors at differentplatform layers. We use platform theory, and in particular theories of multi-sided platforms(MSPs), to focus on key technical and business aspects of today’s industry. We use an MSP-aware layered model of the ecosystem to explore several recent and impending innovations inthe ecosystem that have been naively conflated with the global Internet, illuminate their dif-ferences, and describe how regulators could use our model to more rigorously consider them.Finally, to illustrate its potential as a baseline for future research, we briefly consider how thismodel can help scope consistent policy discourse of three open questions: specialized services,minimum quality regulations (“the dirt road” problem), and structural separation.

1 Introduction

A significant challenge in developing regulatory theory to support communications policy is thehighly dynamic technology and business practices in the evolving Internet. Traditional regula-tory theory in the telecommunications sector relied on simple conceptions of technology, such ascopper pairs carrying telephone service to homes. Innovative uses of that copper pair (e.g. DSL)and advanced technologies such as hybrid fiber-coax (HFC), fiber and wireless, have led to defini-tional confusion, litigation, and a dauntingly complex, poorly understood networked ecosystem.Recent trends toward convergence of virtually all communications services using the Internet Pro-tocol (IP), both in public and private networks, render the complexity and ambiguity even worse.

The goal of this paper is to present a model of communications technology and industry prac-tice that is general enough to survive current rates of innovation and evolution, and stable enoughto support relevant regulatory theory. We draw on two fundamental elements of today’s telecom-munications ecosystem: the use of layered platforms to implement functionality desired by eitherusers or providers; and interconnection between actors at different platform layers to construct alarger fabric. The concepts of platforms, layering, and interconnection are not new; but our modelcombines them in a way that captures both stable and dynamic aspects of the technology andbusiness practices of the ecosystem, while abstracting away details that confuse more than clarifyto regulatory debates.

Section 2 describes a taxonomy of platforms classified across two dimensions: how they areconstructed (one or multiple firms) and how they are used (only by the platform owners or by

1

others). The Internet and its larger ecosystem is actually many layers of platforms that span allfour combinations of these two dimensions. We augment this classification with a well-acceptedmodel of contemporary interconnection patterns among ISPs, and find that the resulting picturebrings some clarity to differences among several recent Internet-related service offerings and theirregulatory implications. The highly dynamic nature of the industry implies that technical detailsof various platform layers will change faster than regulatory debate can possibly resolve, but ourtwo-dimensional platform matrix allow reasoning that is independent of these technical details:whether they are internal or industry platforms, and whether they are single or multi-firm plat-forms. Equipped with this taxonomy, we discuss two platform layers of the ecosystem that seemparticularly durable: the global Internet and the single-firm IP platform. A third emerging plat-form innovation – the multi-firm IP platform – is likely to also become common, which raisessignificant challenges for regulators. Section 3 explores how a platform-aware model can usefullydistinguish among services running on different IP-based platforms, as well as the possible behav-ior of those platform owners, in a way that offers a cleaner derivation and clearer interpretation oftheir regulatory implications. Our underlying premise is that regulation attached to stable layersis likely to be more durable, and Section 4 considers how the framework of platforms we havedeveloped can help scope consistent policy and regulation of the Internet. We study three pol-icy questions of interest in a platform context: specialized services, minimum quality regulations(“the dirt road” problem), and structural separation Section 5 offer some concluding thoughts andsuggests future directions of research and debate.

2 Platforms: their nature and layered structure in the Internet

A platform is a technology providing a set of service capabilities on top of which many differentproducts can be developed and deployed. Operating systems (e.g., Microsoft Windows, iOS, An-droid) are platforms – they support a wide range of applications and services, often provided bymany third-party providers. The owner of a successful platform may acquire powerful advan-tages in the related industrial ecosystem. Factors that influence the ability to create and control asuccessful platform include the dynamics of gaining and holding market share, network external-ities and control of intellectual property.

In the context of industrial structure, Gawer [12] described three classes of platform: internal,supply chain and industry. Internal platforms are developed and exploited internal to a firm toreduce cost and improve flexibility and time to market for product development. Examples ofinternal platforms are often found in manufacturing industries such as automotive and electronics.Supply chain platforms are produced and exploited by a set of firms, linked in collaborations andalliances. An industry platform is produced by one or a set of firms and then made available toother entities (complementors) to develop products on top of that platform.

To clarify their role in our model, Figure 1 depicts four types of platforms (and examples ofeach) along two dimensions: construction and use. The upper row corresponds to Gawer’s in-dustry platform, although we distinguish between those constructed by one vs. multiple firms.The right column generalizes Gawer’s supply chain platform to a multi-firm platform, to includecases where firms are peers or partners. Gawer’s analysis still applies to multi-firm platforms,i.e., relationships among firms producing the platform influence which firms may have sufficientmarket power to define and regulate the platform, or even destabilize it in the process of trying tomaximize profit.

2

!"#$%&'()*'"#+,-.)/'0%123)*'

4"563#&7'8#+)3"+7'9":)3-3;'I7'D5EF'3G&)'D5A'

D#.&)#1%'*,%BC()*'0%123)*'

4F3D5'3G&)'D5'0%123)*A'

J-&K''

DLMJ!ENO'4PQ.&)#1%A'

DLEPNL8R'4E3''.6&':)&1.3)-'

I3#-.),:B3#K''

S#&'()*' !&G&)1%'()*-'

Figure 1: Two important aspects of a platform: how it is con-structed and who its users are. The global Internet is an ex-ample of an external or industry platform produced by manyfirms. The upper left box (an industry platform produced byone firm) is not common in the Internet context, but we pre-dict it will become more common in the future, with profoundimplications for any theory of broadband regulation.

For example, Microsoft Windows is an in-dustry platform, defined and developed bya single firm that controls it through intel-lectual property licensing and other meth-ods to limit appropriation of the platformby competitors. The Internet as we knowit today is a multi-firm industry platform,constructed by multiple firms interconnect-ing their networked assets based on (mostly)open standards. One of the most successfulindustry platforms in history as measured bythe range of innovative applications built bycomplementors using it, the Internet and itslarger ecosystem is actually many layers ofplatforms that span all four quadrants of Fig-ure 1.

The Web is also a multi-firm platform,implemented at a layer above the basic In-ternet transport capability, and applications using this higher-level web platform can themselvesserve as a platforms for other application developers. The Web, like the Internet, is defined bystandards, mostly set by the World Wide Web Consortium (W3C) hosted at MIT, with many ac-tors involved in both setting the standards and implementing them in web servers and browsers.On top of the Web platform one can find further examples of platforms. Facebook is a single-firmindustry platform for complementors who build on top of it. In contrast to the Web platform,which is largely characterized by open standards, the Facebook platform is dictated by its owner,shaped only by any pressures from their complementors and users that they choose to respect.While there have been fears that a powerful player might try to distort a web standard using themarket share of their browser, the need for interoperability among tools (firms) seems to haveprevented serious fragmentation of the Web standards.1

The repeating platform structure of the Internet also appears below the network layer, down tothe physical layer of routers and fibers. Internet service providers provision routers and fiber in or-der to build and maintain their parts of the Internet. But these technologies also enable individualproviders to construct a stack of internal platforms to support a range of services by the firm, or tocreate additional industry platforms for complementors. In network architecture language, theserepeating layers may multiplex, i.e., simultaneously support, many higher-level services. Longdistance fibers are multiplexed by using different colors of light (lambdas) to carry different datastreams. These data streams are often multiplexed using a technology called Multi-Protocol LabelSwitching (MPLS), which breaks the data stream into packets. The most common format for thesepackets is the Internet Protocol (IP), which results in an IP-based platform. From the perspectiveof each ISP, their IP platform is a single-firm platform, while the Internet is a multi-firm platformthat happens to use the same protocol. At the same time, lower layers can serve as both an internalplatform and an industry platform, i.e., the owner of a fiber infrastructure can both use internallyand sell to others a fiber, a lambda, a share of an MPLS service, a range of IP-based services, oraccess to the public Internet.

1For a range of commentary on powerful actors manipulating key web standards, see for examplehttp://www.internetnews.com/dev-news/article.php/83051/Web+Standards+Group+Criticizes+IE+5.htm (1999) orhttp://developers.slashdot.org/story/12/08/10/2152236/microsoft-picks-another-web-standards-fight (2012).

3

N+1$ N+1$

Pla(orm$layer$N$

Lower$pla(orm$layer$N11$

Industry$pla(orm$

Internal$pla(orm$

Complementor$services$

Higher1level$internal$services$$

Lower$layer$

internal$pla(orm$

(a) Platforms sitting on lower layer platforms that can provideboth an industry and internal platform to services above.

Lambdas'

Fibers'

MPLS'

Single2firm'IP'pla6orm'

Global'Internet'

“The'web”'

Facebook'

VoIP' IPTV'

(b) In the Internet ecosystem, fiber owners may be able to cre-ate multiple platform layers, each one serving as an industryplatform as well as an internal platform for the development ofhigher-level services.

Figure 2: Illustration of platform layers and their role in the Internet ecosystem

The significance of these layers inspire the repeating platform layer dimension of our model.Figure 2(a) illustrates how each layer sits on a lower-layer platform that supports a range of in-ternal as well as complementor services; Figure 2(b) instantiates an example set of such platformlayers in the Internet ecosystem. Layered models have been used for at least a decade to explainthe structure of network technology, and to reason about regulatory treatment [19, 23, 24]. Morerecently, Jordan [15] argued that a critical demarcation point is between the Internet Protocol layer(and below) and those layers above the Internet layer (the application or complementor layers).He believes this delineation would not only recover some meaning to the telecommunications vs. in-formation service terminology in the U.S. Telecommunications Act, but would promote innovationat higher layers in the industry by limiting regulation to the lower layers to prevent vertical fore-closure. Our model differs from his in that we articulate the distinction between the multi-firmindustry Internet platform and the IP platform below the Internet; our goal is to avoid simplifica-tions that mask aspects of layering that are relevant to regulatory analysis.

One of our objectives is to identify specific platform layers that are likely to be stable over time,because we argue that stable layers should be the focus of regulatory attention. The physical layerin this repeating layering has a durable character; its creation is typically capital-intensive and itsevolution constrained accordingly. However, technology evolution and dynamism bring tremen-dous variability to the intermediate lower layers of this stack, resulting in different intermediateplatforms, so any attempt to attach regulatory burdens to a specific technical implementation ofa layer runs the risk of rapid irrelevance. For example, in contrast to the MPLS in Figure 2(b),the HFC infrastructure of a modern cable access system implements multiplexing using platformlayers based on 6Mhz channel allocations and a protocol called DOCSIS. Historically contingentdetails of this cascade of platforms will not matter to a general and durable model. In particular,a durable model will be able to identify key features that merit different regulatory treatment ofdifferent layers, regardless of the technical details of a given layer at a given time. Such a modelwill be particularly helpful in the analysis of imminent, economically inevitable platform innova-tions about to shatter our already cracking models of communication regulation. The first suchinnovation we explore is the “single-firm converged IP platform”, often naively conflated with thepublic Internet.

4

2.1 The single-firm IP platform

Most Internet Service Providers have historically provided many services in addition to the publicInternet. These services, such as telephone and cable television, create additional revenue oppor-tunities, perhaps with higher margins than classic Internet service. Today, a common technicalapproach to providing these other services is for a firm to build a single-firm IP platform, a con-verged service layer based on the Internet protocols (IP), over which the infrastructure owner canoffer its own voice (VoIP) or video (IPTV) services, as well as to provide an industry platform suchas the public Internet. The use of the single-firm IP platform can support enhanced service quali-ties (e.g., QoS) that may not be permitted on the Internet platform, due to regulatory restrictions.The term converged captures the idea that this layer brings together disparate lower-level technolo-gies under one service interface. The IP platform is a particularly powerful and successful servicebecause the Internet protocols were designed to support exactly such an objective: to provide alayer that implements a uniform service interface on top of a variety of lower level technologiesand platform interfaces.

This new reality changes the context for reasoning about the role and necessity of regulation.Under the assumption that the major purpose of the IP platform was to implement the public In-ternet, regulatory proposals that tried to constrain the relationship between the ISP and the com-plementors (e.g., ”network neutrality” regulations) considered only the industry platform thatprovided the global Internet. Proponents of network neutrality (or ”reasonable network man-agement,” as the FCC has termed it [8]), are concerned with the potential chilling effect on com-plementors if the Internet provider favored (in pricing or performance) their own higher-levelservices over products and services from complementors. But the regulatory proposition is morecomplex if the owner of the single-firm IP platform can also sell consumer-facing services on topof this internal platform instead of or as well as the global Internet.

Global&interconnected&Internet&

Comcast&single3firm&IP&pla7orm&

AS&

The&Web&

Xbox&TV&

OTT&TV&

Figure 3: Delivery of Comcast Xfinity video over their con-verged IP platform, in contrast to over the top (OTT) videobeing delivered by competitors over the Internet.

We cannot easily dismiss these multipleuses of the single-firm IP platform, since theymay be the only way to recover the costsof operating the underlying infrastructure ina competitive market. For example, accessISPs today offer what is sometimes calledthe triple play over their access technology –voice, video and Internet; revenues from allof these services cover the costs of the infras-tructure.2

As an example of a service offered overa single-firm IP platform, Comcast recentlyintroduced the Xfinity Xbox IPTV service,which carries on-demand television content over IP to a customer’s Xbox for display on an at-tached monitor television (Figure 3) [16]. Comcast described this service as a consumer-facingservice running on top of their single-firm IP platform, not the public Internet platform, and thusasserted that it was not covered by the obligations against discrimination imposed by the FCC’sOpen Internet Report and Order. Therefore they exempted the usage attributed to this servicefrom the monthly quota on their Internet service. Critics argued that the Xfinity Xbox serviceshould be viewed as operating on the global Internet platform, and therefore Comcast should(under the Open Internet Report and Order) treat it equally to other video services and include it

2The FCC concurred with this assessment of the economics of broadband access in their ruling on local franchisingof cable competitors [7, para 51].

5

Single'firm+IP+pla0orm+

Global+Internet+

Xbox+TV+

DOCSIS+

Cable+channels+

Tradi@onal+digital+Cable+TV+

Today+

Single'firm+IP+pla0orm+

Global+Internet+

Xbox+TV+

DOCSIS+

Cable+channels+

IP'based+TV+

Future+

AS+

AS+

Figure 4: Most cable TV providers will migrate TV onto their single-firm converged IP platform. No IPinteroperation (i.e. over a multi-firm platform) is required to deliver IPTV.

in the cap.3

This debate will intensify as all of cable television service will eventually move to this single-firm IP platform layer model, i.e., TV will become IPTV (figure 4), which does not imply it is run-ning as just another application on the global Internet. New services can be expected to emerge aswell–services that have no prior regulatory history as cable TV and telephony do. A challenge forregulators is to assess the implications, both positive and negative, of the emergence of consumer-facing services offered by facilities owners over their single-firm (internal) IP platform.

Further complicating the analysis, a firm could convert their internal platform to an external(industry) platform, by allowing complementors access (i.e., direct interconnection) to this plat-form (thus moving from the lower left to the upper left quadrant of the matrix in Figure 1). Thispossibility reveals a fundamental question: if both the single-firm IP platform and the global In-ternet are supporting complementors, perhaps some of the same complementors, in what respectsare they different platforms? Alternatively, two firms might agree to interconnect their internal IPplatforms for a specific purpose, shifting along the other dimension of the platform matrix (fromthe lower left to the lower right quadrant of Figure 1).

Before we explore the implications of these two types of interconnection, we summarize thefour important aspects of the ecosystem our model has captured this far. First, the ecosystemhas many layers, sometimes recursive (IP on top of another IP layer), and with technical detailsthat change rapidly enough that a stable and general model of regulation will have to ignorethese technical details. Second, one can reason independent of these technical details by firstcharacterizing platforms along the two dimensions in our matrix: whether they are internal orindustry platforms, and whether they are single or multi-firm platforms. Third, the emergingsingle-firm IP platform is likely to be stable and persistent, which means our model has at leasttwo durable layers: the global Internet and the single-firm IP platform. Fourth, we anticipate theemergence of a third stable platform in the IP ecosystem: a multi-firm IP platform, which mayserve both as an industry or an internal platform. We next examine issues related to such multi-

3For one perspective on this debate see http://gigaom.com/2012/03/27/the-technical-and-legal-realities-of-comcasts-xbox-cap-spat/

6

Tier%1%provider%

Tier%1%provider%

Tier%1%provider%

Tier%2%ISP%

Smaller%ISP%

Smaller%ISP%

Smaller%ISP%

Smaller%ISP%

Smaller%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Transit%

Peering%

(a) Physical representation of Internet interconnection.

Tier%1%provider%

Tier%1%provider%

Tier%1%provider%

Tier%2%ISP%

Smaller%ISP%

Smaller%ISP%

Smaller%ISP%

Smaller%ISP%

Smaller%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Tier%2%ISP%

Transit%Peering%

CDN%

(b) The addition of a content delivery network (CDN) to themix of interconnections.

Figure 5: Both the multi-firm production of the global Internet and the interconnection of complementorssuch as CDNs may manifest as physical links in a diagram of the Internet.

firm platforms, of which the Internet itself is the most recognized one.

2.2 From single-firm platform to multi-firm platform

We have described the global Internet as an industry platform composed of an interconnectedmesh of many single-firm IP platforms. Interconnection is the process, evolving for the last severaldecades, of building a multi-firm platform out of the single-firm IP platform of each ISP.

Internet interconnection patterns have evolved over the past decades. In the past, while mosttraffic used to travel up to a Tier 1 provider on its way to its destination, over the years moretraffic began to pass over a dense mesh of Tier 2 ISPs rather than transiting major Tier 1 providers.Figure 5(a) illustrates the resulting mesh of Tier 2 peering. The more interesting recent evolution,from the perspective of this paper, is the emergence of large content delivery networks (CDNs),which attach to the various regions of the Internet to facilitate the efficient delivery of content, asillustrated in Figure 5(b). This figure emphasizes the physical topology of interconnection, notdistinguishing CDNs from ISPs in terms of their role in the ecosystem.

Illustrating interconnection in terms of layers of platforms, as in Figure 6(a), clarifies this dis-tinction. In this figure, content delivery networks operate at a layer above the global Internetplatform, because their role in the ecosystem is not general interconnection among firms that con-stitute the global Internet, but rather as a service to content providers. A CDN is a complementorof the ISP platforms to which it connects, as well as being another platform in its own right, actingas a delivery enhancement mechanism for a range of higher-level services. Content-related com-plementors (e.g. a Content Delivery Network or CDN and its customers) are superficially similarto ISPs: they have AS numbers, interconnect with other ISPs, etc. But interconnection between aCDN and a broadband access provider crosses platform layers, between complementor and In-ternet. In other words, some physical interconnection involves the construction of the multi-firmplatform, and some interconnection represents a use of that platform.

Spulber and Yoo [22] identify five sorts of interconnection that can be found in the Internet:retail (our end-users), wholesale, interconnection (our multi-firm production of the Internet plat-form), platform (the compementors) and unbundled access (unbundling at the physical layer).They stress the difference between the retail and platform interconnection, although they discussthe platform interconnection mostly in terms of access to TCP/IP, rather than as physical intercon-nection. They recognize, as we stress, that two sets of links may be identical in terms of physical

7

Converged)IP)pla/orm)

AS)

Global)interconnected)Internet)

Converged)IP)pla/orm)

AS)

The)Web) Content)delivery)network)

Firm)1) Firm)2)

Converged)IP)pla/orm)

AS)

Firm)3)

(a) Platform model of the multi-firm Internet, created by in-terconnecting Autonomous Systems (ASes), i.e., parts of theglobal Internet operated by one firm.

Single'firm+IP+pla0orm+

AS+

Global+interconnected+Internet+

Interconnected+VoIP+

Single'firm+IP+pla0orm+

AS+Private+IP+ Private+IP+

Global+private+IP+network+

The+Web+

Firm+1+ Firm+2+

(b) Both the global Internet and a multi-firm internal IP net-work built by interconnecting the platforms of multiple firms.

Figure 6: The converged IP platform supports two global platforms: (a) the public Internet and (b) a privateIP network. Lines in (b) represent physical interconnections as well as connections between platform layers.

properties but different in terms of the economic relationships. The paper presents an economicanalysis of the implications of using regulation to manage each of these forms of interconnection.

2.3 Alternative multi-firm IP platforms

A more recent development in this evolutionary trajectory is that firms that have built the multi-firm global Internet by interconnecting their single firm platforms may replicate this approach toproduce other multi-firm IP (but not Internet) platforms. The initial role seen in practice for sucha network is as an internal platform, serving the firms that create it. For example, some firmsthat offer VoIP are now creating a second connected mesh of private IP services over which tointerconnect VoIP calls. Figure 6(b) illustrates the creation of both the global Internet (an industryplatform) and a multi-firm internal IP platform by the interconnection of single-firm platformsacross multiple firms. Just as a single-firm IP platform can serve as an industry platform as wellas an internal platform, a multi-firm IP platform could be offered as an industry platform to com-plementors. Such an interconnected IP platform would be an alternative (to the Internet) industryplatform for third-party complementors to reach consumers instead of the current global Internet,perhaps offering better quality of service or security. An infrastructure provider might marketsuch a platform to large application providers such as Facebook as a more effective way to reachconsumers, e.g., see figure 7. As with single-firm IP platforms, the inevitable question for theregulator is how to view companies building, interconnecting, and selling services on their ownprivate IP-based platforms, logically if not physically separate from the global Internet platform.

The FCC’s Open Internet Report and Order [8] generates some urgency to this question ofwhen an alternative service offering is similar enough to the global Internet that it should berequired to comply with the order. The current view expressed (although perhaps not clearly)in this report is that the answer depends on the reach and character of the alternative service: ifit reaches essentially as far as the public Internet, then it should comply with the Order, whichwould impose rules about discrimination and blocking [8, para. 47,112]. However, the multi-firm IP platform that carries VoIP services for the facilities owner (e.g., it is currently an internalplatform) will likely end up with a global scope, and is not currently subject to neutrality rules. Itis not clear why reach is the correct criterion for regulatory classification of a service.

In this section we have mapped various IP-based platforms into our matrix of options for thecreation and exploitation of platforms. We have shown examples of an IP platform in each of thefour quadrants, and explained how a platform-aware understanding of interconnection can shed

8

light on otherwise confusing innovations in the ecosystem. Next we introduce another relevantaspect of platform theory that is essential to understanding the economics of today’s Internetinfrastructure: the multi-sided platform, or MSP.

2.4 Multi-sided platforms (MSPs)

Single'firm+IP+pla0orm+

Public+Internet+

Global+interconnected+Internet+

Single'firm+IP+pla0orm+

Public+Internet+Private+IP+ Private+IP+

Mul:'firm+IP+network+

The+Web+

Firm+1+ Firm+2+

The+Web+

Facebook+

“open”+non'

discriminatory+pla0orm+

Private+pla0orm+

with+op:on+for+discrimina:on.+

Improved+service+

Figure 7: Both the global Internet and a multi-firm IP net-work acting as alternative industry platforms for consumer-facing services.

A Multi-Sided Platform (MSP) is a specialcase of a multi-sided market. A multi-sidedmarket is a circumstance in which a providerhas several classes of mutually dependentparticipants. Hagiu [14] provides a discus-sion of different definitions of MSPs. He alsofollows the tradition of using the term cus-tomers to describe the classes that are associ-ated with the market. However we will usethe term participant to stress the point that theclasses of participants that associate with theprovider of the market may not all be pay-ing to use the platform. The classic exam-ple of a multi-sided market is a dating club,where both men and women must attend forthe club to be successful, and a club owner might offer men and women different prices to enticethem both to come, because of a perception that men and women would attach different values tothe experience. An MSP, by analogy, is a platform that requires several classes of participants topartake in order for the platform to be successful.

By this definition, the global Internet is an MSP, with retail users as one class of participantsand complementors as another. Complementors may participate in a number of ways: provid-ing software, web content, gaming, music sharing, or other services. Complementors and usersare mutually dependent; for any activity involving both classes, there is a balance between theirroles. Wikipedia is a complementor on the global Internet platform that provides a service (itselfa single-firm industry platform) to enable users (contributors) to produce value. For peer-to-peerfile sharing, the complementor only produces and perhaps maintains the sharing software, andusers create value by using it. For commercial content creation and distribution, the user is asomewhat passive consumer and the complementor drives production.

While the global Internet taken as a whole is arguably an MSP, from both a business and regu-latory perspective the unit of concern is usually the firm (or a class of firms, like broadband accessproviders), not a multi-firm platform. This distinction raises the question of whether it is legiti-mate to view an individual firm as an MSP. According to Hagiu’s definition [14], an MSP’s classesof participants must have a direct association with the platform provider. Historically, a comple-mentor attached to the Internet at one or a few points, but served all users on the Internet fromthose points. Most access ISPs transited packets to and from their peers, but did not have explicitrelationships or even necessarily know about any complementors. Even today, most complemen-tors who provide services using network-attached servers (e.g. Web servers or game servers) payto attach to the Internet, but since much of the Internet interconnects via revenue-neutral peering,no revenues from complementors flow into the access ISP. Retail consumers of broadband access(“eyeballs”) cover the entire costs of the access ISP.

This oversimplified description of the Internet would not classify most individual access ISPsas MSPs, since the complementors as a class of participants do not have a direct association with

9

the access ISP. There are some exceptions to this case, of course. For example, ISPs that provideWi-Fi hot spot access for free, financed by advertising revenues, seem like MSPs, with users andadvertisers forming the two classes of participants. But with respect to complementors as a classof participants, an explicit association between complementors and access ISPs is increasinglycommon today, moving individual access ISPs toward an MSP role in the ecosystem. Most no-tably, large content providers such as Netflix and Google, as well as CDNs that serve content onbehalf of many other content providers, often negotiate business relationships that result in di-rect interconnection to access ISPs, as illustrated in Figure 5(b). As another example of a directassociation between access ISP and complementor, HBOgo (the Internet version of HBO) relies onthe access ISP to verify that a specific user has the right access credentials based on that customeralso subscribing to HBO via their cable TV service. ESPN requires that access ISPs pay a fee toESPN in order that its users have access to the Internet ESPN service, called ESPN3. As describedin Section 2.2, physical interconnection can reflect either the production of the multi-firm Internetplatform or the connection of a complementor to an IP platform; supporting this latter type ofinterconnection turns an access ISP into an MSP.

3 Regulatory implications of a multi-sided platform model

Analyses of multi-sided markets usually focus on differential pricing [17], but much of the con-cern today with regulation of access ISPs relates to other forms of discriminatory behavior, e.g.,blocking or variable quality of delivery of content from complementors. In an analysis of regula-tory implications of MSP behavior, Evans [10] concludes that a potential regulator must be cau-tious about determining predatory pricing in an MSP, as the price on one side may be efficientlyset above or below cost, as part of the overall pricing analysis. Generalizing Evans’ argumentsuggests that not just pricing but all forms of potential discrimination should be evaluated withregard to all classes of participants. Several researchers have explicitly argued that certain formsof discrimination can enhance the overall health of an MSP [9, 13, 2].

For example, the Apple app store (a single-firm industry platform) is an MSP, with two classesof participants: the application designers and the purchasers. Apple discriminates among applica-tions (the complementors). While this business strategy may restrict competition and innovationat the level of the complementors, a thorough MSP analysis should examine the impacts on allplatform participants. Discrimination with respect to applications arguably makes the platformmore attractive to users, by giving them confidence that undesirable or risky applications havebeen removed from the system. This outcome might in turn increase market share and makethe platform more attractive to complementors. (Note the large difference between an MSP exer-cising some degree of discrimination with respect to a complementor class and a powerful MSPblocking a popular application, perhaps as an exercise in rent-seeking or protection of a verticallyintegrated competitor. Apple removed the Google maps app from the iPhone in favor of their ownmaps, triggering a significant consumer backlash.)

In the remainder of this section we explore how multi-sided platform theory would influencediscourse on different types of discrimination by platform owners against complementors andretail users.

3.1 Discrimination with respect to Internet complementors

Analysis of the consequences of a particular form of discrimination will depend on the higher-level goal that drives the call for non-discrimination. Why are advocates and regulators con-

10

cerned about non-discrimination on the Internet? One goal is fostering innovation, i.e., non-discriminatory access protects complementors [1]. Others have argued that non-discriminatoryaccess fosters the production of public and social goods [11]. But as Evans’ argument suggests,and Apple’s ecosystem demonstrates, some forms of discrimination with respect to the comple-mentors might result in a platform that appeals to more users, and thus fosters more productionof downstream goods. The potential complexity of the MSP suggests that protection of comple-mentors and promotion of downstream goods (including public and social goods) merit distinctregulatory analysis with respect to discrimination.

With respect to complementors, ISPs impose some coarse discrimination behavior today, viavolume discounts, which favor large, entrenched complementors over new entrants. ISPs can alsoimpose discrimination using deep packet inspection (DPI) to distinguish and prioritize traffic, orlimiting a class of application via software on an attached device.4 ISPs could also offer discrim-inatory terms for physical interconnection to complementor networks. The MSP model suggeststhat all forms of potential discrimination imposed by the platform owner on their complemen-tors should use one regulatory logic, as opposed to viewing physical interconnection through onelens, DPI through another, and so on. The concern is the business relationship, not the technicalapproach to discrimination.

We are certainly not the first to argue that a multi-sided analysis is appropriate to understandthe relationship between complementors and users. Bourreau, Kourandi and Valletti [3] use theconcept of the two-sided market to model the relationship between content producers and retailcustomers. Their framing is similar to ours, in that they are interested not just in pricing, but in theinfluence of discrimination (or not) among the complementors (in this specific model providersof free content supported by advertising) on consumers, and the resulting implications for theoverall ecosystem. They conclude that allowing ISPs to offer enhanced QoS to complementors inexchange for payment is welfare-enhancing. They also conclude from their model that it may bein the interest of an ISP to degrade the basic best-effort service, so that regulatory attention maybe required. Their work models two competing ISPs; they cite earlier work that looks at a singlemonopoly ISP. Sidak [20] has a similar analysis in which he concludes that it is beneficial to allowISPs to charge content providers for enhanced QoS.

3.2 Peering disputes and pricing analyses

In the last five years we have seen increasingly public debates about whether peering betweennetworks should be revenue-neutral (without financial settlements between peers) or whetherpaid peering should reflect relative costs to ISPs of carrying traffic, or even reflect manifestations ofrelative bargaining power. The MSP model suggests that regulatory scrutiny of peering disputesshould distinguish between two different circumstances: disputes among the firms making up theplatform, or discrimination by platform operators on complementors on the platform. A peeringdispute between an ISP and a complementor should be subject to the same sort of analysis as anyother form of discrimination imposed by an ISP on a complementor.

With respect to peering disputes among firms that interconnect to make up the Internet, suchevents are characteristic of what can happen in any multi-firm platform. As described by Gawer [12],in any multi-firm platform a powerful actor can dominate the creation of the platform, or poten-tially even disrupt it by predatory behavior. In particular, the argument is much weaker thatan ISP is situated in an MSP when it negotiates terms for peering with other ISPs than when itnegotiates with complementors who need access to users.

4For example, ATT used a feature in the iPhone software to block use of the FaceTime app for certain data plans. Seehttp://arstechnica.com/apple/2013/01/att-continues-chipping-at-facetime-over-cellular-policy/.

11

Peering disputes with complementors have gained more recent attention, given the increasingnumber and size of complementors seeking direct associations with access ISPs. Some accessISPs have questioned the proper balance of pricing and discrimination between a specific classof complementor (content distribution networks) and users, arguing that distributors of contentshould pay for the right to connect to the ISP’s customers [18]. This situation is a classic example ofa pricing analysis in an MSP; previously, one class of participant, the end-consumer, was coveringthe costs of the access ISP; an alternative is to extract some of that revenue from another classof participant. Although this move might signal the exercise of market power by the ISP, Evansargues [10] that the presence of market power does not affect the analysis of efficient pricing inan MSP. Extraction of rents is an issue with market power, efficient pricing is not. So, ignoringthe issue of market power, what are the implications of shifting the balance of payments betweenusers and complementors of the access ISP? It is unreasonable to assume that the ISP could exerciseprecise first-degree price discrimination and charge each complementor a price that reflects theirwillingness to pay, given both its impracticality as well as the likely reaction from the market andthe regulators. A more likely approach is for the access ISP to charge complementors per bit ofdata generated by the service, which might drive out of the market complementors with low orunproven value. On the other hand, if this shift significantly drove down the price charged tousers, more users might enter the market.

We know of no fundamental principle that resolves this balance; it calls for empirical datasuch as cost and price elasticity. However, the complexity of the MSP landscape implies thatISPs contemplating a shift in their negotiating position about interconnection with complementorsshould not consider the shift only as a new source of revenues. Even non-disciminatory pricescharged to complementors, if high enough, could chill innovation. Lower prices may benefitinnovation and public good creation more than the total absence of discrimination, especially inthe complementor market.

3.3 Discrimination with respect to retail users

As a preliminary to the rules promulgated in the Open Internet Report and Order, and to frametheir regulatory objectives to “ensure that broadband networks are widely deployed, open, afford-able, and accessible to all consumers”, the FCC published four principles [6]:

To encourage broadband deployment and preserve and promote the open and inter-connected nature of the public Internet, consumers are entitled to

• access the lawful Internet content of their choice.• run applications and use services of their choice, subject to the needs of law en-

forcement.• connect their choice of legal devices that do not harm the network.• competition among network providers, application and service providers, and

content providers.

The first two principles, without using the language or framing of an MSP, in fact do a two-sided analysis, looking at the consequences for the consumer (one class of participant) of discrim-ination on another class (complementors–providers of content and applications). The third prin-ciple directly protects the user from one form of discrimination, and the fourth principles offersa bit of wishful hoping that competition happen. But the only direct protection of the user fromdiscrimination is that the consumers can attach devices of their choice. There is no requirementthat all potential users be served, or served at the same price.

12

In fact, ISPs today do discriminate with respect to what consumers can do. One common lim-itation is that residential users are not allowed to attach a “server” (device) to their network. Thisrestriction might fall within the “reasonable network management” exception on a cable systemwith limited upstream capacity, but when Google imposed the same restriction on their KansasCity service with abundant upstream capacity, they received some criticism [21]. Another viewis that the the limited protection given to the consumer by the four principles was intentionallyconstructed to allow ISPs to stratify their users into classes, e.g., residential vs. business.

With respect to pricing, most wireline ISPs do not differentially charge users based on whatthey are doing. Rather, they offer either flat pricing or usage caps. These billing models are a formof discrimination: some customers with low usage support those with higher usage. The benefitto this form of discrimination is that for any individual user, the incremental cost of trying somenew service from a complementor, or engaging in an activity that generates mostly a public good(e.g. the user does not appropriate the benefit) is zero. In contrast, wireless plans that imposeusage caps and tiers may inhibit experimentation and participation in activities where the valuethat accrues to the user is (perceived to be) low.

3.4 MSP behavior in more mature markets

Another consequence of applying MSP theory relates to saturation in one side of the market.According to MSP theory, when platforms compete (as with IP vs. another protocol suite) in agrowing market, innovation by complementors brings value to the platform owner in the form ofnetwork effects that lead to success of the platform. This dynamic applied in the early days of theInternet. But once the platform (as an interface or standard) has achieved market dominance, andespecially if the pool of users saturates, as it seems to be doing for land-line broadband access, thelogic of the MSP may cease to explain the benefit to the platform provider of more complementors(i.e., apps). Indeed, with current wireline pricing models that bring little or no additional revenueswith additional usage, new complementors that drive increased usage bring negative value to theplatform provider. Alternatively, usage tiers may encourage investment in capacity, and thus theaddition of new complementors, but at the same time discourage the use of these complementsby cost-conscious users.

4 Shaping consistent regulation based on layered MSP model

Our repeating layer model captures the idea that a given actor (e.g., an ISP) may produce manyplatforms as part of their system design. We have focused on three platforms that appear (orare likely) to be stable in the ecosystem: the global Internet, the single-firm IP platform, and amulti-firm IP platform. We have also explored how the theory of multi-sided platforms can shedfurther light on the behavioral influences of and on platform owners, complementors, and users.Since regulation attached to stable layers would likely be more durable, we now consider how theframework of platforms we have developed can help scope consistent policy and regulation of theInternet. We examine three open issues and what light platform theory sheds on them: one short-term regulatory quagmire, one medium-term policy challenge, and one long-term philosophicaldebate.

4.1 Consistent regulation of services on different IP-based platforms

Regulators, when looking at the structure of the market, may look at the overall behavior of firms,or at the desired qualities of a platform. Put differently, when should a regulator consider a plat-

13

form in isolation, as opposed to considering a firm’s treatment of several platforms. MSP theorysuggests an answer to this question: platform layers merit a unified analysis if they serve the sameclasses of participants. Whether the platform is single-sided and serves one class of participants,or an MSP serves multiple classes, if two platforms can serve the same classes then they should beconsidered jointly. As a specific example, if an access ISP offers its single-firm IP platform as an al-ternative industry platform for (at least some) complementors to reach consumers, the single-firmand global Internet platforms are alternative means to serve the same classes of participants, andmerit joint consideration. In contrast, an access ISP might use its single-firm MPLS platform layerto offer an “Ethernet-over-MPLS” service to allow a multi-site enterprise to connect its varioussites; this type of enterprise constitutes a different class of customer, and thus this platform meritsindependent consideration from an MSP IP platform.

Note that joint consideration of platforms does not imply equal treatment. The single-firm IPplatform as a industry platform for complementors imposes extra burdens on potential comple-mentors: it serves only complementors that arrange to connect directly to it. If (as we describedearlier) multiple access ISPs interconnected their single-firm IP platforms to build a second, multi-firm IP industry platform that served the same set of user classes (complementors and consumers)with the same implications for those classes of participants, it would seem the two platformswould be more likely to deserve similar regulatory treatment. Some entanglement potentiallyarises between platforms if they serve some but not all of the same participant classes. As in theXfinity Xbox service (Section 2.1), an access ISP could use its internal single-firm IP platform tooffer a video service to consumers (e.g., it could act as a reseller of video content over that plat-form), which arguably competes with a video service offered by a third-party complementor overthe global IP platform. The services over the two platforms (single-firm IP and the Internet) havethe retail user in common as a class of participants, but they do not involve the same classes ofparticipants: with carrier IPTV, the ISP is acting as a reseller of video programming, not as an MSP.Any entanglement of regulatory consideration would derive from the fact that services over bothplatforms offer competing and similar offerings to one class of participants.

We observe that the FCC tried to address this situation by using the term “specialized services”in the Open Internet Report & Order [8]; a specialized service is any service that is not burdenedwith the obligations of the report and order. Like many other terms introduced to describe newbehaviors that do not fit into the prevailing model of industry behavior, the term “specialized ser-vices” brings little clarity to what is happening. The FCC specifically identifies carrier IPTV as aspecialized service (e.g., IPTV running over the single-firm IP platform of the provider, althoughthey do not use this terminology or framing), but the more general (and not yet resolved) questionis when to classify a service as a specialized service. The FCC’s current position is that the dis-tinction is determined by the set of machines reachable via the service [8, para. 47]. Comcast, indescribing their Xfinity IPTV service, stressed that the IPTV and Internet service use different logi-cal slices of Comcast’s bandwidth capacity.5 These sorts of distinctions do not seem useful as a testof whether a service is running on the internal IP platform or on the global Internet service. MSPtheory suggests that a more fundamental and promising starting point for classification of special-ized services is the degree to which the two services have common classes of participants. Thus,carrier-provided IPTV is a video reseller service that competes with services that complementorsoffer over the global Internet. In an MSP-based framework, the basic question is whether this typeof competition between platforms should warrant any regulatory attention, not the details of howcapacity is allocated or routing is restricted.

5See http://corporate.comcast.com/comcast-voices/the-facts-about-xfinity-tv-and-xbox-360-comcast-is-not-prioritizing

14

4.2 Regulating the platform itself rather than the platform provider

The previous section described an approach to regulation in which the regulator imposes require-ments on a firm, perhaps across several platforms, to limit its behavior with respect to variousclasses of its users, such as complementors. An alternative is to regulate a specific platform, withthe goal of sufficiently satisfying high-level social goals (e.g., high performance, low cost, comple-mentor innovation) so that regulation of other platforms of the firm is not required. Such platformregulation will attempt to ensure that the player with power (the platform provider) is workingfor the overall good of the ecosystem, but this abstract policy statement leaves much unansweredabout the right approach. We examine potential regulation of the global Internet platform as aspecific case study.

The higher-level question remains: what goal might warrant regulatory intervention? If thegoal is to drive healthy innovation, then if the global Internet is “good enough” to stimulate com-plementor innovation, it matters less what the provider does with its single-firm IP platform. Inother words, the higher performing and lower cost the global Internet offering, the lower the reg-ulatory interest should be in activities on the single-firm (or multi-firm) IP layer. (This approachraises the question of how to define the minimum quality standards, which others have alreadyrecognized as a potential anchor of future regulation [4, 5].)

The traditional discourse on network neutrality (and specifically in the U.S., the FCC OpenInternet principles [6]) approaches this question primarily by constraints on the Internet offeringitself. But the alternative platforms, particularly the single and multi-firm IP platforms, increasethe degrees of freedom both for an ISP and for the regulator. Any theory of regulation that arguesfor limitation on the activities of the facilities owner must take into account the degrees of freedomthat the platform owner has. For example, imagine that a platform operator (broadband provider)allocates a share of the IP platform as an alternate multi-firm industry platform, which is IP-based,but not interconnected with the public Internet. On that share, the platform operator allows third-party complementors to offer consumer-facing services, perhaps with superior qualities comparedto the public Internet (e.g. QoS), or may curate the available applications like the Apple app store.A theory of regulation should be able to explain why this business behavior is pro-innovationor anti-competitive, pro-consumer or not. The MSP analysis can illuminate debate of these sortsof questions. It may reveal, given the degrees of freedom to both actors, that a better regulatoryapproach is to offer incentives to the ISP to improve the global Internet platform, as we describeabove, rather than constraining its operation.

4.3 Structural separation

Finally, we use our multi-sided platform model to describe and compare options for a more radicalapproach to industry regulation. In platform terms, structural separation is a requirement that theowner of some platform be required to use it only for third- party complementors, but not as aninternal platform for deploying higher-level services. Structural separation is a extreme form offacilities unbundling; in simple unbundling, the owner of the platform must make it available toany comer on non-discriminatory terms, but is also allowed to use it as an internal platform onthe same terms.

As a real-world example, the British regulator has imposed structural separation on the phys-ical layer platform of British Telecom (BT). BT was required to establish a separate firm, Open-Reach, to own and operate the actual copper pairs. In this paper we have not emphasized theplatform represented by the physical layer, but it is indeed another durable layer and thus a rea-sonable target of regulation. However, several aspects of the physical layer make it a challenging

15

target of structural separation regulation. First, this layer is very dependent on particular tech-nology choices. It is difficult to imagine a technology-neutral form of structural separation at thephysical layer. Second, using copper pairs as an example, there is no simple way to share a cop-per pair between two complementors. A given consumer will pick among competing providersof higher-level services, but then must obtain all services from that one provider. The resultingmarket structure would seem to limit competition among complementors.

Imposing a separation point at a higher-level platform would allow more flexibility in theresulting market structure. In particular, separation at a layer that allows multiplexing amongmultiple complementors (e.g., a packet-switched platform) could allow multiple complementorsto offer services to the same customer. The packet level could be the single-firm IP platform, or alower-level platform such as an ATM platform. Our emphasis on finding durable platform layersaround which to regulate implies that structural separation of an intermediate layer (between thephysical and the single-firm IP platform) would be most likely to succeed if the platform whereseparation was imposed was defined in a technology-independent way, and thus could apply toa range of alternative technologies that will be found in those layers over time.

We are not advocating structural separation; just framing it in terms of our model. But thetools of MSP and multi-platform analysis could help to answer the question in any debate overstructural separation: what are the implications for the platform owner and the larger ecosystemif the owner were precluded from using it as an internal platform.

5 Conclusions

We have presented a model that characterizes both the layered platform architecture of today’scommunications technology, and aspects of interconnection among firms to produce what wehave identified as the global Internet and its connected complementors. We have identified threespecific IP-based platforms as important: the single-firm IP platform, the global Internet, and amulti-firm private IP platform. These platform layers will be more stable fixtures of the infras-tructure than other layers beneath or above them, and thus a focus on these specific layers willhelp produce policy that is durable in the face of evolving business and technology. Distinguish-ing the three is necessary in any realistic regulatory or business analysis. With respect to issuesrelated to interconnection, we argue that distinguishing between the role of multi-firm productionand complementor interconnection is a fundamental criterion. Peering among firms that make upthe global Internet raises different issues than interconnection between a complementor and theInternet.

Our model suggests opportunities for discrimination on multi-sided platforms, both with re-spect to users and to complementors. These include pricing discrimination (some of which we seetoday, such as volume discounts and flat rate pricing), discrimination in interconnection policy,and manipulation of traffic flows based on their characteristics, including which users or whichcomplementors generate the traffic. In trying to understand which of these behaviors should betreated together rather than separately, we argue that with respect to a particular platform, allforms of discrimination with respect to a given class of participant (e.g., a complementor) shouldbe analyzed using the same framework, as opposed to looking at (for example) interconnectionusing one framework and deep packet inspection using another. With respect to the questionof whether behavior on different platforms should be evaluated independently or in common,we argue that a starting point is to determine the extent the two platforms serve the same set ofparticipants.

16

References

[1] Barbara van Schewick. Internet Architecture and Innovation. MIT Press, 2010.

[2] Kevin Boudreau and Andrei Hagiu. Platform rules: Multi-sided platforms as regulators. InPlatforms, Markets and Innovation. Edward Elgar, 2011.

[3] Marc Bourreau, Frago Kourandi, and Tommaso Valletti. Net neutrality with competing inter-net platforms, 2012. Available at SSRN: http://ssrn.com/abstract=2061571orhttp://dx.doi.org/10.2139/ssrn.2061571.

[4] Tim Brennan. Net Neutrality or Minimum Quality Standards: Network Effects vs.Market Power Justifications, 2010. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1622226.

[5] kc claffy and David. Clark. Workshop on Internet Economics (WIE2012) Report. ACM SIG-COMM Computer Communication Review (CCR), July 2013.

[6] Federal Communications Commission. Fcc 05-151, policy statement, 2005. In the matter ofAppropriate Framework for Broadband Access to the Internet over Wireline Facilities, CCDocket No. 02-33, etc., http://hraunfoss.fcc.gov/edocs_public/attachmatch/FCC-05-151A1.doc.

[7] Federal Communications Commission. In the matter of implementation of section 621(a)(1)of the cable communications policy act of 1984 as amended by the cable television consumerprotection and competition act of 1992, 2005. MB Docket No. 05-311, FCC 06-180.

[8] Federal Communications Commission. Report and order, preserving the open inter-net, fcc10-201, 2010. http://hraunfoss.fcc.gov/edocs_public/attachmatch/FCC-10-201A1_Rcd.pdf.

[9] Thomas Eisenmann, Geoffrey Parker, and Marshall Van Alstyne. Open platforms: How,when and why. In Platforms, Markets and Innovation. Edward Elgar, 2011.

[10] David A. Evans. The antitrust economics of multi-sided platform markets. Yale Journal onRegulation, 20(2), 2003.

[11] Brett M. Frischmann. Infrastructure: The Social Value of Shared Resources. Oxford UniversityPress, 2012.

[12] Annabelle Gawer. Platform dynamics and strategies: From products to services. In Platforms,Markets and Innovation. Edward Elgar, 2011.

[13] Shane Greenstein. Open platform development and the commercial internet. In Platforms,Markets and Innovation. Edward Elgar, 2011.

[14] Andrei Hagiu and Julian Wright. Multi-sided platforms. Technical Report 12-024, HarvardBusiness School, 2011. available at http://hbswk.hbs.edu/item/6681.html?

[15] Scott Jordan. A layered network approach to net neutrality. International Journal of Communi-cation, 1:427–460, 2007.

[16] Rob Frieden. The mixed blessing of a deregulatory endpoint for the public switched tele-phone network. Telecommunications Policy, July 2012.

17

[17] Jean-Charles Rochet and Jean Tirole. Two-sided markets: A progress report. The RANDJournal of Economics, 37(3):645–667, 2006. http://www.jstor.org/stable/25046265.

[18] Sean Buckley. Cogent and Orange France fight over interconnection issues. Fierce Telecom,2011.

[19] Douglas Sicker and Joshua Mindel. Refinements of a layered model for telecommunicationspolicy. Journal on Telecommunications and High Technology Law, pages 69–94, 2002.

[20] J. Gregory Sidak. What is the network neutrality debate really about? International Journal ofCommunication, 1, 2007. http://ijoc.org/index.php/ijoc/article/view/177/95.

[21] Ryan Singel. Now that it’s in the broadband game, google flip-flops on network neutrality,2013. http://www.wired.com/threatlevel/2013/07/google-neutrality/.

[22] Daniel F. Spulber and Christopher S. Yoo. Network regulation: The many faces of access.Journal of Competition Law and Economics, 1(4):635–678, 2005.

[23] Philip J. Weiser. Law and information platforms. Journal on Telecommunications and HighTechnology Law, pages 1–36, 2002.

[24] Kevin Werbach. A layered model for internet policy. Journal on Telecommunications and HighTechnology Law, pages 37–68, 2002.

18